Classify Radio Signals - 2D Spectrograms into 4 Categories ( Squiggle, Narrowband, Noise and Narrowbanddrd) using Convolutional Neural Network.
This is a 2D Spectrogram image data of Radio Signals. Initially, it was initially a time-series data, captured by SETI(Search for extraterrestrial intelligence) institute, which was later converted to 2D Spectogram image data. The 2D Spectogram enables us to treat this as a image-classification problem.
The dataset can be downloaded from https://drive.google.com/drive/folders/1FaDxc0yEh7T7mY1v1LEOS4NoUs2mu7pJ?usp=sharing .
- Keras ImageDataGenerator function is used as a data generator function.
- Image Augmentation of Horizontal Flip is used.
This project uses 3 Convolutional Networks (With Batch Normalization, Activation Function Relu and Dropout Regularization) , 1 Dense Layer with Dropout and the 4-class Output Layer with Softmax Activaion.
To increase the convergence with gettin accurate results, Keras Exponential Decay Function is used.
The model aquires 75.13% accuracy on the validation set.
Graph of the Loss and Accuracy over epoch :
